Design and Rationale of HiLo: A Pragmatic, Randomized Trial of Phosphate Management for Patients Receiving Maintenance Hemodialysis

Abstract

Rationale & objective: Hyperphosphatemia is a risk factor for poor clinical outcomes in patients with kidney failure receiving maintenance dialysis. Opinion-based clinical practice guidelines recommend the use of phosphate binders and dietary phosphate restriction to lower serum phosphate levels toward the normal range in patients receiving maintenance dialysis, but the benefits of these approaches and the optimal serum phosphate target have not been tested in randomized trials. It is also unknown if aggressive treatment that achieves unnecessarily low serum phosphate levels worsens outcomes.

Study design: Multicenter, pragmatic, cluster-randomized clinical trial.

Setting & participants: HiLo will randomize 80-120 dialysis facilities operated by DaVita Inc and the University of Utah to enroll 4,400 patients undergoing 3-times-weekly, in-center hemodialysis.

Intervention: Phosphate binder prescriptions and dietary recommendations to achieve the "Hi" serum phosphate target (≥6.5 mg/dL) or the "Lo" serum phosphate target (<5.5 mg/dL).

Outcomes: Primary outcome: Hierarchical composite outcome of all-cause mortality and all-cause hospitalization. Main secondary outcomes: Individual components of the primary outcome.

Results: The trial is currently enrolling.

Limitations: HiLo will not adjudicate causes of hospitalizations or mortality and does not protocolize use of specific phosphate binder classes.

Conclusions: HiLo aims to address an important clinical question while more generally advancing methods for pragmatic clinical trials in nephrology by introducing multiple innovative features including stakeholder engagement in the study design, liberal eligibility criteria, use of electronic informed consent, engagement of dietitians to implement the interventions in real-world practice, leveraging electronic health records to eliminate dedicated study visits, remote monitoring of serum phosphate separation between trial arms, and use of a novel hierarchical composite outcome.

Trial registration: Registered at ClinicalTrials.gov with study number NCT04095039.

Keywords: End-stage kidney disease (ESKD); cluster-randomized trial; dialysis clinics; hemodialysis; hierarchical composite outcome; hospitalization; hyperphosphatemia; mortality; pragmatic clinical trial; serum phosphate; study design.

Conflict of interest statement

Financial Disclosure: Dr Wolf reports honoraria from Akebia and Ardelyx. The remaining authors declare that they have no relevant financial interests.

Copyright © 2020 The Authors. Published by Elsevier Inc. All rights reserved.

Figures

Figure 1.

Equipoise for HiLo. With multiple factors in favor of and against more aggressive reduction of serum phosphate levels, there is clinical equipoise to conduct a randomized trial of strict versus liberal management of hyperphosphatemia in patients with kidney failure undergoing hemodialysis. Abbreviation: CVD, cardiovascular disease. Created with BioRender.com.

Figure 2.

Differences between pragmatic and traditional explanatory clinical trials. Pragmatic clinical trials differ from traditional explanatory trials across multiple domains. Here, we demonstrate how the current pragmatic design of HiLo would differ if it were conducted as a traditional explanatory trial. Created with BioRender.com.

Figure 3.

Cluster versus individual randomization in HiLo. To maintain the fidelity of the intervention using a pragmatic approach, HiLo employs cluster randomization in which entire dialysis facilities are randomized instead of randomizing patients individually. To balance the sizes of the Hi and Lo arms, HiLo will stratify the cluster randomization by facility type (Davita Inc or University of Utah) and by facility size (less or more than the median; latter represented as outlined groups). Created with BioRender.com.

Figure 4.

Analysis of the primary hierarchical composite outcome of HiLo. The primary outcome of the HiLo trial is a hierarchical composite outcome of time to all-cause mortality and number of hospitalizations. In a pairwise manner, the outcomes for each patient randomized to the Lo and Hi groups are compared. Follow-up duration for each pairwise comparison is defined by the earliest occurrence of death or a censoring event (eg, loss to follow-up or end of study) in either comparator patient. The pairs are first compared on the basis of survival to determine the “winner” who survived longer. If both patients in a pairwise comparison survive throughout the duration of the shared follow-up period, the winner is the patient who had fewer hospitalizations during follow-up. Note that the shared follow-up period is the only time relevant to a given pairwise comparison. As a corollary, an individual patient’s follow-up time will vary across pairwise comparisons depending on the duration of follow-up in their comparators. The final 2 scenarios denote the continuation of pairwise comparisons for Lo patient 1 with all n remaining Hi patients (second from bottom) and all n remaining Lo patients with all n Hi patients (bottom) until all comparisons of all Hi versus all Lo patients are complete. After all pairwise comparisons are complete, the total scores in the 2 arms are tallied and tested for significant differences. Created with BioRender.com.

Figure 5.

Sample size and power simulations for HiLo. Because conventional power calculation methods cannot be applied to hierarchical composite outcomes, HiLo used simulated datasets based on data from DaVita Inc and the TiME trial to determine power and sample size. The figure lists the key assumptions that were used to randomly generate 5,000 iterations of simulated study databases for each of 2 different study compositions: 80 clusters of 55 patients each and 120 clusters of 36 patients each. Then, 2 analytic approaches were used to analyze the simulated datasets: one parametric (mixed model) and one nonparametric (Wilcoxon rank-sum method). All 4 approaches yielded power of 85% or higher. Abbreviation: ICC, intraclass correlation coefficient. Created with BioRender.com.